Method tracing and system-wide process sampling

v4.2

Features

• Java Method Tracing and Latency Profiling
  • async-profiler#1421: Latency profiling
  • async-profiler#1435: Allow wildcards in Instrument profiling engine
  • async-profiler#1499: --trace option with per-method latency threshold
• System-wide process sampling on Linux
  • async-profiler#1411: --proc option to record profiler.ProcessSample events
• VMStructs stack walker by default
  • async-profiler#1539: Use VMStructs stack walking mode by default
  • async-profiler#1537: Support comptask and vtable features
  • async-profiler#1517: Use JavaFrameAnchor to find top Java frame
  • async-profiler#1449: Special handling of prologue and epilogue of compiled methods

Improvements

• async-profiler#1475: Add CPUTimeSample event support to jfrconv
• async-profiler#1414: Per-thread flamegraph option in JFR heatmap converter
• async-profiler#1526: Expose JfrReader dictionary that maps osThreadId to javaThreadId
• async-profiler#1448: Thread name in OpenTelemetry output
• async-profiler#1413: Add time_nanos and duration_nanos to OTLP profiles
• async-profiler#1450: Unwind dylib stubs as empty frames on macOS
• async-profiler#1416: Add synthetic symbols for Mach-O stubs/trampolines
• Allow cross-compilation for 32-bit platforms

Bug fixes

• async-profiler#1515: Fix UnsatisfiedLinkError when tmpdir is set to a relative path
• async-profiler#1500: Detect if calloc calls malloc for nativemem profiling
• async-profiler#1427: Re-implement SafeAccess crash protection
• async-profiler#1417: Two wall-clock profilers interfere with each other

Project Infrastructure

• async-profiler#1527: GHA: replace macos-13 with macos-15-intel
• async-profiler#1510: Add option to retry tests
• async-profiler#1508: Add more GHA jobs to cover JDK versions on ARM
• async-profiler#1502: Fix job dependencies between integration tests and builds
• async-profiler#1466: Add Liberica JDK on Alpaquita Linux to the CI
• Made integration tests more stable overall

OpenTelemetry support and improved native memory profiler

v4.1

Features

• Experimental support for the OpenTelemetry profiling signal
  • async-profiler#1188: OTLP output format and dumpOtlp Java API
  • async-profiler#1336: JFR to OTLP converter
• JDK 25 support
  • async-profiler#1222: Update VMStructs for JDK 25
• Productize native memory profiling
  • async-profiler#1193: Full nativemem support on macOS
  • async-profiler#1254: Fixed Nativemem tests on Alpine
  • async-profiler#1269: Native memory profiling now works with jemalloc
  • async-profiler#1323: nativemem shows allocations inside async-profiler itself

Improvements

• async-profiler#1174: Detect JVM in non-Java application and attach to it
• async-profiler#1223: Native API to add custom events in JFR recording
• async-profiler#1259: --all option to collect all possible events simultaneously
• async-profiler#1286: Record which CPU a sample was taken on
• async-profiler#1299: Skip last 10% allocations for leak detection
• async-profiler#1300: Allow profiling kprobes/uprobes with --fdtransfer
• async-profiler#1366: Rewrite jfrconv executable to shell
• async-profiler#1400: Unwind checksum and digest intrinsics on ARM64
• async-profiler#1357, async-profiler#1389: VMStructs-based stack unwinding for alloc and nativemem profiling

Bug fixes

• async-profiler#1251: --ttsp option does not work on Alpine
• async-profiler#1264: Guard hook installation with dlopen/dlclose
• async-profiler#1319: SIGSEGV in PerfEvents::walk
• async-profiler#1350: Disable JFR OldObjectSample event in jfrsync mode
• async-profiler#1358: Do not dereference jmethodIDs on JDK 26
• async-profiler#1374: Correctly check if profiler is preloaded
• async-profiler#1380: Workaround clang type promotion bug
• async-profiler#1387: JFR writer crashes when using cstack=vmx
• async-profiler#1393: Improve stack walking termination logic: no endless unknown frames
• Stack unwinding fixes for ARM64

Project Infrastructure

• async-profiler#1129: Command-line option to filter tests
• async-profiler#1262: Include asprof.h in async-profiler release package
• async-profiler#1271: Release additional binaries with debug symbols
• async-profiler#1274: Add Corretto 8 to the test matrix
• async-profiler#1246, async-profiler#1226: Run tests on Amazon Linux and Alpine Linux
• async-profiler#1360: Auto-generated clang-tidy review comments
• async-profiler#1373: Save all generated test logs for debug purposes
• Fixed flaky tests (async-profiler#1282, async-profiler#1307, async-profiler#1376)

Heatmaps and Native memory profiling

v4.0

Features

• async-profiler#895, async-profiler#905: jfrconv binary and numerous converter enhancements
• async-profiler#944: Interactive Heatmap
• async-profiler#1064: Native memory leak profiler
• async-profiler#1002: An option to display instruction addresses
• async-profiler#1007: Optimize wall clock profiling
• async-profiler#1073: Productize VMStructs-based stack walker: --cstack vm/vmx
• async-profiler#1169: C API for accessing thread-local profiling context

Improvements

• async-profiler#923: Support JDK 23+
• async-profiler#952: Solve musl and glibc compatibility issues; link libstdc++ statically
• async-profiler#955: --libpath option to specify path to libasyncProfiler.so in a container
• async-profiler#1018: --grain converter option to coarsen flame graphs
• async-profiler#1046: --nostop option to continue profiling outside --begin/--end window
• async-profiler#1178: --inverted option to flip flame graphs vertically
• async-profiler#1009: Allows collecting allocation and live object traces at the same time
• async-profiler#925: An option to accumulate JFR events in memory instead of flushing to a file
• async-profiler#929: Load symbols from debuginfod cache
• async-profiler#982: Sample contended locks by overflowing interval bucket
• async-profiler#993: Filter native frames in allocation profile
• async-profiler#896: FlameGraph: Alt+Click to remove stacks
• async-profiler#1097: FlameGraph: N/Shift+N to navigate through search results
• async-profiler#1182: Retain by-thread grouping when reversing FlameGraph
• async-profiler#1167: Log when no samples are collected
• async-profiler#1044: Fall back to ctimer for CPU profiling when perf_events are unavailable
• async-profiler#1068: Count missed samples when estimating total CPU time in ctimer mode
• async-profiler#1142: Use counter-timer register for timestamps on ARM64
• async-profiler#1123: Support clock=tsc without a JVM
• async-profiler#1070: Demangle Rust v0 symbols
• async-profiler#1007: Use ExecutionSample event for CPU profiling and WallClockSample for Wall clock profiling
• async-profiler#1011: Obtain can_generate_sampled_object_alloc_events JVMTI capability only when needed
• async-profiler#1013: Intercept java.util.concurrent locks more efficiently
• async-profiler#759: Discover available profiling signal automatically
• async-profiler#884: Record event timestamps early
• async-profiler#885: Print error message if JVM fails to load libasyncProfiler
• async-profiler#892: Resolve tracepoint id in asprof
• Suppress dynamic attach warning on JDK 21+

Bug fixes

• async-profiler#1143: Crash on macOS when using thread filter
• async-profiler#1125: Fixed parsing concurrently loaded libraries
• async-profiler#1095: jfr print fails when a recording has empty pools
• async-profiler#1084: Fixed Logging related races
• async-profiler#1074: Parse both .rela.dyn and .rela.plt sections
• async-profiler#1003: Support both tracefs and debugfs for kernel tracepoints
• async-profiler#986: Profiling output respects loglevel
• async-profiler#981: Avoid JVM crash by deleting JNI refs after GetMethodDeclaringClass
• async-profiler#934: Fix crash on Zing in a native thread
• async-profiler#843: Fix race between parsing and concurrent unloading of shared libraries
• async-profiler#1147, async-profiler#1151: Deadlocks with jemalloc and tcmalloc profilers
• Stack walking fixes for ARM64
• Converter fixes for jfrsync profiles
• Fixed parsing non-PIC executables and shared objects with non-standard section layout
• Fixed recursion in pthread_create when using native profiling API
• Fixed crashes on Alpine when profiling native apps
• Fixed warnings with -Xcheck:jni
• Fixed "Unsupported JVM" on OpenJ9 JDK 21
• Fixed DefineClass crash on OpenJ9
• JfrReader should handle custom events properly
• Handle truncated JFRs

Project Infrastructure

• Restructure and update documentation
• Implement test framework; add new integration tests
• Unit test framework for C++ code
• Run CI on all supported platforms
• Test multiple JDK versions in CI
• Add GHA to validate license headers
• Add Markdown checker and formatter
• Add Issue and Pull Request templates
• Add Contributing Guidelines and Code of Conduct
• Run static analyzer and fix found issues (async-profiler#1034, async-profiler#1039, async-profiler#1049, async-profiler#1051, async-profiler#1098)
• Provide Dockerfile for building async-profiler release packages
• Publish nightly builds automatically

v3.0.2

Another SAP interim release that contains latest patches and works on JDK23.

v3.0.1

SAP interim release that works with JDK23.

Binary launcher and AsyncGetCallTrace replacement

v3.0

Features

• async-profiler#724: Binary launcher asprof
• async-profiler#751: Profile non-Java processes
• async-profiler#795: AsyncGetCallTrace replacement
• async-profiler#719: Classify execution samples into categories in JFR converter
• async-profiler#855: ctimer mode for accurate profiling without perf_events
• async-profiler#740: Profile CPU + Wall clock together
• async-profiler#736: Show targets of vtable/itable calls
• async-profiler#777: Show JIT compilation task
• async-profiler#644: RISC-V port
• async-profiler#770: LoongArch64 port

Improvements

• async-profiler#733: Make the same libasyncProfiler work with both glibc and musl
• async-profiler#734: Support raw PMU event descriptors
• async-profiler#759: Configure alternative profiling signal
• async-profiler#761: Parse dynamic linking structures
• async-profiler#723: --clock option to select JFR timestamp source
• async-profiler#750: --jfrsync may specify a list of JFR events
• async-profiler#849: Parse concatenated multi-chunk JFRs
• async-profiler#833: Time-to-safepoint JFR event
• async-profiler#832: Normalize names of hidden classes / lambdas
• async-profiler#864: Reduce size of HTML Flame Graph
• async-profiler#783: Shutdown asprof gracefully on SIGTERM
• Better demangling of C++ and Rust symbols
• DWARF unwinding for ARM64
• JfrReader can parse in-memory buffer
• Support custom events in JfrReader
• An option to read JFR file by chunks
• Record GCHeapSummary events in JFR

Bug fixes

• Workaround macOS crashes in SafeFetch
• Fixed attach to OpenJ9 on macOS
• Support UseCompressedObjectHeaders aka Lilliput
• Fixed allocation profiling on JDK 20.0.x
• Fixed context-switches profiling
• Prefer ObjectSampler to TLAB hooks for allocation profiling
• Improved accuracy of ObjectSampler in --total mode
• Make Flame Graph status line and search results always visible
• loop and timeout options did not work in some modes
• Restart interrupted poll/epoll_wait syscalls
• Fixed stack unwinding issues on ARM64
• Workaround for stale jmethodIDs
• Calculate ELF base address correctly
• Do not dump redundant threads in a JFR chunk
• check action prints result to a file
• Annotate JFR unit types with @ContentType

Maintenance release

v2.5.1

Bug fixes

• Prevent early unloading of libasyncProfiler.so
• Read kernel symbols only for perf_events
• Escape backslashes in flame graphs
• Avoid duplicate categories in jfrsync mode
• Fixed stack overflow in RedefineClasses
• Fixed deadlock when flushing JFR

Improvements

• Support OpenJDK C++ Interpreter (aka Zero)
• Allow reading incomplete JFR recordings

macOS/ARM64 port and many JFR improvements

v2.5

Features

• macOS/ARM64 (aka Apple M1) port
• PPC64LE port (contributed by @ghaug)
• Profile low-privileged processes with perf_events (contributed by @Jongy)
• Raw PMU events; kprobes & uprobes
• Dump results in the middle of profiling session
• Chunked JFR; support JFR files larger than 2 GB
• Integrate async-profiler events with JDK Flight Recordings

Improvements

• Use RDTSC for JFR timestamps when possible
• Show line numbers and bci in Flame Graphs
• jfr2flame can produce Allocation and Lock flame graphs
• Flame Graph title depends on the event and --total
• Include profiler logs and native library list in JFR output
• Lock profiling no longer requires JVM symbols
• Better container support
• Native function profiler can count the specified argument
• An option to group threads by scheduling policy
• An option to prepend library name to native symbols

Notes

• macOS build is provided as a fat binary that works both on x86-64 and ARM64
• 32-bit binaries are no longer shipped. It is still possible to build them from sources
• Dropped JDK 6 support (may still work though)

Maintenance release

v1.8.7

Bug fixes

• Workaround for JDK-8173361
• Backported fix for "Accept timed out" exception

Maintenance release

v1.8.6

Improvements

• log=none option to suppress warnings about missing JVM symbols
• Sign macOS binaries

Bug fixes

• Workaround for JDK-8212160